Re: [PATCH] printk: Fix print_time length computation when no buffer is given

[Joe Perches <joe@perches.com>]

On Thu, 2012-12-06 at 18:09 +0100, Sylvain Munaut wrote:
> The "%5lu" part of the sprintf only guarantees a minimum length,
> not a maximum one. This patch should make it correct for any
> possible timestamp.
> 
> This problem dates back from the printk conversion to records AFAICT.
> 
> I've tested this locally both by verifying 'dmesg' no longer
> segfaulted and also by checking that before this patch the
> buffer was overrun by exactly the number of char corresponding to
> extra digits in the timestamps.
> 
> Cc: Kay Sievers <kay@vrfy.org>
> Cc: stable@kernel.org
> Signed-off-by: Sylvain Munaut <s.munaut@whatever-company.com>
> ---
>  kernel/printk.c |   13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/printk.c b/kernel/printk.c
> index 66a2ea3..2ceceea 100644
> --- a/kernel/printk.c
> +++ b/kernel/printk.c
> @@ -847,8 +847,17 @@ static size_t print_time(u64 ts, char *buf)
>  	if (!printk_time)
>  		return 0;
>  
> -	if (!buf)
> -		return 15;
> +	if (!buf) {
> +		size_t len = 15;
> +		u64 ts_limit = (u64)100000 * (u64)1000000000;
> +
> +		while ((ts > ts_limit) && (len < 21)) {
> +			len++;
> +			ts_limit = 10 * ts_limit;
> +		}

Perhaps use snprintf for sizing something like below?
(untested/uncompiled)

static size_t print_time(u64 ts, char *buf)
{
	unsigned long rem_nsec;
	char *tmpbuf = buf;
	int size = 15;			/* expected size of non-null buf */
	char local_buf[1];

	if (!printk_time)
		return 0;

	if (!buf) {			/* emit to a dummy buf instead */
		tmpbuf = local_buf;
		size = sizeof(local_buf);
	}
	
	rem_nsec = do_div(ts, 1000000000);

	return snprintf(tmpbuf, size, "[%5lu.%06lu] ",
			(unsigned long)ts, rem_nsec / 1000);
}

It's perhaps unfortunate that print_time doesn't take a
a buffer and a size_t of buffer.